CN110175297A - Personalized every member's model in feeding - Google Patents
Personalized every member's model in feeding Download PDFInfo
- Publication number
- CN110175297A CN110175297A CN201910121550.1A CN201910121550A CN110175297A CN 110175297 A CN110175297 A CN 110175297A CN 201910121550 A CN201910121550 A CN 201910121550A CN 110175297 A CN110175297 A CN 110175297A
- Authority
- CN
- China
- Prior art keywords
- user
- model
- entry
- machine learning
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000000694 effects Effects 0.000 claims abstract description 39
- 230000006855 networking Effects 0.000 claims description 63
- 238000010801 machine learning Methods 0.000 claims description 60
- 238000000034 method Methods 0.000 claims description 24
- 230000009471 action Effects 0.000 abstract description 6
- 230000008080 stochastic effect Effects 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 23
- 230000015654 memory Effects 0.000 description 20
- 230000006399 behavior Effects 0.000 description 15
- 238000004891 communication Methods 0.000 description 15
- 238000004422 calculation algorithm Methods 0.000 description 12
- 238000012549 training Methods 0.000 description 12
- 230000005540 biological transmission Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- 230000004044 response Effects 0.000 description 10
- 230000003993 interaction Effects 0.000 description 9
- 238000001514 detection method Methods 0.000 description 7
- 230000008878 coupling Effects 0.000 description 6
- 238000010168 coupling process Methods 0.000 description 6
- 238000005859 coupling reaction Methods 0.000 description 6
- 230000002452 interceptive effect Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000033001 locomotion Effects 0.000 description 5
- 238000007726 management method Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 230000003287 optical effect Effects 0.000 description 5
- 230000009021 linear effect Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000000712 assembly Effects 0.000 description 3
- 238000000429 assembly Methods 0.000 description 3
- 230000003542 behavioural effect Effects 0.000 description 3
- 230000001413 cellular effect Effects 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 241000700605 Viruses Species 0.000 description 2
- 230000008676 import Effects 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012797 qualification Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 230000036772 blood pressure Effects 0.000 description 1
- 230000036760 body temperature Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000010267 cellular communication Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000008921 facial expression Effects 0.000 description 1
- 230000037406 food intake Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 231100001261 hazardous Toxicity 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000007620 mathematical function Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- DZCCLNYLUGNUKQ-UHFFFAOYSA-N n-(4-nitrosophenyl)hydroxylamine Chemical compound ONC1=CC=C(N=O)C=C1 DZCCLNYLUGNUKQ-UHFFFAOYSA-N 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 230000008261 resistance mechanism Effects 0.000 description 1
- 210000001525 retina Anatomy 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2457—Query processing with adaptation to user needs
- G06F16/24578—Query processing with adaptation to user needs using ranking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
In the exemplary embodiment, using GLMix model, the GLMix model models the observer of feeding entry and actor.This allows to consider in the case where not introducing deviation the stochastic effects of individual viewer and actor.In addition, in the exemplary embodiment, by using three models (and then it is combined) rather than single GLMix model so that prediction/recommendation is more acurrate.Each of these models all have different granularity and size.World model can model the similitude between user property (for example, from members profiles or activity history) and entry attributes.Every viewer's model can model the user property and activity history for feeding the actor in entry.Every actor's model can to feeding entry viewer user property and activity history model.Therefore, every actor's model may rely on about following information: what kind of viewer interacts with the entry acted on by specific action person and how to interact.
Description
Technical field
The technical issues of being encountered when the present disclosure relates generally to provide personalized prediction on the computer network.More specifically,
This disclosure relates to the use of the broad sense additivity melange effect machine learning model for personalized prediction.
Background technique
Social networking service be people be used to establish with other people social networks or social networks in line platform.In recent years
Come, social networking service has caught on, to provide a user feeding (feed), wherein can be to the use of login service
Update or interested entry is presented in family.For example, feeding may include it is boosted, for the social networks connection of user
Changed the instruction of operation etc..Feeding can also include the interested article of user, or the social activity because of them and user
One or more of network connection (for example, article that friend writes) has some connections, or because they are linked to user's
Area-of-interest, as social networking service identified (for example, article has been identified as the sense in its user profiles in user
On the theme in interest region).
Shown for social networking service it is determined that in feeding which of many potential entries and it
The sequence that should be shown may be with challenge.This is usually via the one or more realized by social networking service
Algorithm is handled, to show for potential entry to be selected and arranged.However, these algorithms are based on using single
A machine learning model determines that user (for example, select it, share it, as it etc.) will interact with entry in some way
A possibility that.However, individual machine learning model means that all members share identical weight in the world.Therefore,
Although member can based on themselves data (such as past interaction) receive personalized score the fact, due to
How data influence individual machine learning model from other members calculates score, so score is devious.
Detailed description of the invention
In the figure of attached drawing, some embodiments of this technology are shown by way of example, and not limitation.
Fig. 1 is to show the block diagram of client-server system according to example embodiment.
Fig. 2 is the block diagram shown with the functional unit of the consistent social networking service of some embodiments of the present disclosure, should
The functional unit of social networking service includes the data processing module of referred to here as search engine, is used to generate and offer is searched
The search result of rope inquiry.
Fig. 3 is the block diagram for illustrating in greater detail the application server module of Fig. 2 according to example embodiment.
Fig. 4 is to illustrate in greater detail to show the operation publication sort result engine of Fig. 3 according to example embodiment
Block diagram.
Fig. 5 depicts the machine learning model 412 of Fig. 4 according to example embodiment in more detail.
Fig. 6 is the screenshot capture according to user's feeding 600 including different classes of entry of some example embodiments.
Fig. 7 is the stream for the method 700 using broad sense additivity Mixed effect model shown according to example embodiment
Cheng Tu.
Fig. 8 is to show the block diagram of software architecture according to example embodiment.
Fig. 9 shows the graphical representation of the machine of computer system form according to example embodiment, in the computer
One group of instruction can be executed in system so that machine executes any one or more of process discussed herein.
Specific embodiment
It summarizes
The present disclosure describes the method, system and computer program product that various functions are provided separately and other things.
In the following description, for illustrative purposes, numerous specific details are set forth in order to provide the different embodiments to the disclosure
The thorough understanding of various aspects.It is apparent, however, to one skilled in the art, that can be in no all these tools
The disclosure is practiced in the case where body details.
In history, the model for being ranked up to potential feeding (feed) entry is largely utilized mentions from feeding entry
The text that takes and whole world sequence is derived based on the feature of entity or is recommended.This model another example is generalized linears
Model (GLM).GLM is the generalization of linear regression, allows the response with the model of error distribution in addition to normal distribution
Variable.GLM is by allowing linear model related to response variable via link function and passing through the variance of each measurement of permission
Magnitude be its predicted value function come to linear regression carry out generalization.
Following predictor formula can be used in GLM:
Response of this formula predictions user i to entry j, and xijIt is feature vector, w is coefficient
Vector,The expectation being in response to, and g () is link function.
However, in user or entry rank there is more fine-grained model will potentially lead in the case where data rich
More accurately prediction is caused, because user may more preferably grab the personal preference and entry of entry to the specific attraction of user
It obtains.
In the exemplary embodiment, by introducing member's grade regression coefficient other than the global regression coefficient during GLM is arranged
To provide for preferably capturing the personal preference and entry to entry of user to user in prediction/recommender system
Specific attraction solution.This is referred to as Generalized Linear Mixed Models (GLMix).
In the exemplary embodiment, GLMix model, viewer and actor of the GLMix model to feeding entry are utilized
It is modeled.This allows to consider in the case where not introducing deviation the stochastic effects of individual viewer and actor.
In addition, in the exemplary embodiment, by using three models rather than single GLMix model so that predicting/pushing away
It is more accurate to recommend.Specifically, be not the single GLMix model possessed for viewer and actor with different coefficients, and
It is then to be combined using three individual models.Each of these models all have different granularity and size.Global mould
Type can model the similitude between user property (for example, from members profiles or activity history) and entry attributes.
Every viewer's model can model the activity history and user property for feeding the actor in entry.Every actor's mould
Type can to feeding entry viewer activity history and user property model.Therefore, every actor's model can be according to
Lai Yu on how to and the information that interacts of the entry that is acted on of what kind of viewer and specific action person.
This model is properly termed as broad sense additivity melange effect (GAME) model.For the purpose of this disclosure, potential feeding
The actor of entry is the user acted via the user interface execution of social networking service, which makes the entry quilt
It is considered as potential feeding entry.The example of such movement may include updating members profiles, shared potential feeding entry, happiness
Joyous potential feeding entry, publication article etc..
In potentially feeding entry sequence or the context recommended, this causes with lower component:
World model, the general behavior how capture member interacts with feeding entry
Specific to the model of viewer, there is the parameter (will learn from data) specific to given member, to catch
Obtain member about check feeding entry personal behavior, and
Specific to the model of actor, there is the parameter (will learn from data) specific to actor, with capture
Actor member about to the personal behavior taken action of feeding entry and the feeding item acted on by actor member
The behavior of purpose observer.
It is the description how GAME model realizes rank personalization below.Enable ymjtIndicate whether user m will be upper and lower
Interaction (for example, click) is responded by the binary of the actor j entry acted in literary t, and wherein context generally includes entry and shown
Time and position when showing.qmFor indicating the feature vector of user m comprising the spy extracted from the public profiles of user
Sign, for example, the academic title of member, job function, education history, industry etc. and past feeding interactive history.sjFor table
Show the feature vector of actor j comprising the feature extracted from the public profiles of actor, for example, the academic title of member, work
Make function, education history, industry etc. and past feeding interactive history, including before with the entry that is acted on by actor
The attribute of the user interacted and interaction.Enable xmjtThe global feature vector for indicating (m, j, t) triple, may include being used for
The q of feature level main effectmAnd sj, for the q of the interaction between observer and actor's featuremAnd sjBetween external produce
The feature of product and context.It assume that xmjtIt does not include to be used as feature by member identities (ID), because ID will be with routine
Feature is treated with a certain discrimination.The probability that user m is interacted with the feeding entry acted on by actor j is predicted for using logistic regression
GAME model be:
E[ymjt]=x 'mjtb+s′jαm+q′mβj
Wherein b is global coefficient vector (also referred to as fixed effect coefficient)；And αmAnd βjIt is specific for observer m respectively
With the coefficient vector of actor j.αmAnd βjReferred to as stochastic effects coefficient, capture observer m to different entry features
Personal preference and by actor j to the entry of the attraction role of different members feature.For the past to different items
There are many viewers responded for mesh, this can accurately estimate his or her personal coefficient vector αmAnd provide personalized prediction.
On the other hand, if observer m does not have too many past response data, αmPosterior Mean will be close to zero, and
Model for user m will return back to global fixed effect component x 'mjtb.Identical behavior be suitable for every actor's coefficient to
Measure βj。
Fig. 1 is to show the block diagram of client-server system 100 according to example embodiment.Networked system 102 passes through
Server side functionality is provided from network 104 (for example, internet or wide area network (WAN)) to one or more clients.Fig. 1 shows
The web client 106 (for example, browser) for example executed on each client machine 110 and 112 and program client are gone out
End 108.
Application Programming Interface (API) server 114 and web server 116 are coupled to one or more application server
118 and program and web interface are respectively supplied to one or more application server 118.118 trustship of application server one
Or multiple apply 120.Then, application server 118 is shown to be coupled to one or more database servers 124, is convenient for
Access one or more databases 126.Although showing the application 120 of a part for being used to form networked system 102 in Fig. 1,
It will be appreciated that in an alternative embodiment, application 120 can be formed to be separated and different services from networked system 102
A part.
In addition, although client-server system 100 shown in Fig. 1 uses client-server architecture, this
It is open to be certainly not limited to this framework, and equally application can be found in for example distributed or peer-to-peer architecture system.It is various
It also may be implemented to be independent software program using 120, not necessarily have networked capabilities.
The web interface access that web client 106 is supported via web server 116 is various to apply 120.Similarly, program
The various services and function that the programming interface access that client 108 is provided via API server 114 is provided by application 120.
Fig. 1 also shows the third-party application 128 executed on third-party server 130, has via API service
Programmatic access of the programming interface that device 114 provides to networked system 102.For example, third-party application 128 can use from networking
The information that system 102 is fetched is supported by the one or more features or function on the website of third party's trustship.For example, third party
Website can provide the one or more functions supported by the related application 120 of networked system 102.
In some embodiments, any website being mentioned above may include can be presented in various equipment it is online
Content, the various equipment include but is not limited to desktop PC (PC), laptop computer and mobile device (for example,
Tablet computer, smart phone etc.).In this respect, user can use the disclosure using any one of these equipment
Feature.In some embodiments, mobile device (any client machine 110,112 and third-party server can be used in user
130 can be mobile device) on mobile application access and browse online content, such as it is any disclosed herein online
Content.Mobile Server (for example, API server 114) can be communicated with mobile application and application server 118, to make
The feature of the disclosure can be used on the mobile apparatus.
In some embodiments, networked system 102 may include the functional unit of social networking service.Fig. 2 is to show
With the block diagram of the functional unit of the consistent social networking system 210 of some embodiments of the present disclosure.In some embodiments, it searches
Index, which holds up 216, may reside on application server 118 shown in Fig. 1.It is contemplated, however, that other configurations are also in the disclosure
In range.
As shown in Fig. 2, front end may include Subscriber Interface Module SIM (for example, web server 116) 212, reception is come from
The request of various client computing devices, and response appropriate is transmitted to requesting client equipment.For example, Subscriber Interface Module SIM
212 can receive the request in the form of hypertext transfer protocol (HTTP) request or other API requests based on web.Separately
Outside, the person of may be provided in interacts detection module 213 to detect the various of the different application 120 of member and presentation, service and content
Interaction.As shown in Fig. 2, member's interaction detection module 213 is in member activity and behavior database when detecting specific interactive
Recording interactive in 222, including interactive type and with any metadata of intercorrelation.
It may include one or more various application server modules 214 using logical layer, in conjunction with Subscriber Interface Module SIM
212, generate the various user interfaces (for example, webpage) for the data that there are the various data sources from data Layer to fetch.Some
In embodiment, each application server module 214 is for realizing the various applications 120 and/or clothes provided with social networking service
It is engaged in associated function.
As shown in Fig. 2, data Layer may include several databases 126, such as the profile data for storing profile data
Library 218, the profile data include members profiles' data and the profile number for various tissues (for example, company, school etc.)
According to.It is consistent with some embodiments, when people's first registers become the member of social networking service, it will be prompted to the people's offer
Some personal information, for example, his or her name, the age (for example, date of birth), gender, interest, contact information, local,
Address, the name of spouse and/or kinsfolk, education background (for example, school, profession, preparatory course and/or date of graduation etc.),
Work experience, technical ability, professional association etc..The information is for example stored in profiles database 218.Similarly, when the representative of tissue
When initially registering the tissue to social networking service, the representative can be prompted to provide certain information about the tissue.The information
It can for example be stored in profiles database 218 or another database (not shown).In some embodiments, it can handle
Profile data (for example, from the background or offline) is to generate various derived profile datas.For example, if member provide it is related
The various academic titles and information how long that the member and same tissue or different tissues are held, then the information can be used for pushing away
Break or derives that members profiles' attribute, members profiles' attribute show in the overall qualification rank or specific organization of the member
Qualification rank.In some embodiments, it imports or otherwise accesses from one or more hosted outside data sources
Data can enrich the profile data for member and tissue.For example, financial data can be from one especially for tissue
Or multiple external data sources import, and as a part of organizing-profile.
Once registration, member can invite other members or be invited by other members, to carry out via social networking service
Connection." connection " may be constructed the bilateral agreements carried out by member, so that two members recognize establishment of connection.Similarly,
In some embodiments, member can choose " concern " another member.It is different from connection is established, " following " another member
Usually one-sided operation, and at least in some embodiments, it is not required to the confirmation or approval of member to be traced.When one
When a member follows another member, the member paid close attention to can receive state and update (for example, in activity or content stream
In) or by other message tracked member's publication or related with the various activities that tracked member carries out.Class
As, when member pays close attention to tissue, the qualified reception of the member represents the message of tissue publication or state updates.For example, representing
The individuation data feeding that the message of member's tissue publication of interest or state update will appear in member is (commonly referred to as living
Dynamic stream or content stream) in.Under any circumstance, member and other members or the various associations established with other entities and object
With relationship by storage and maintenance in the socialgram in social graph data library 220.
When member interacts with various applications 120 available via social networking service, service and content, member's
Interactive and behavior (for example, the content checked, selected link or button, the message responded etc.) can be tracked, and
And for example, it can recorde or store the work about member by member activity and behavior database 222 by indicated in Fig. 2
Dynamic and behavior information.Then it can be determined using the action message of the record for search inquiry by search engine 216
Search result.
In some embodiments, database 218,220 and 222 can be merged into the database 126 in Fig. 1.However,
Other configurations are also within the scope of this disclosure.
Although being not shown, in some embodiments, social networking system 210 provides API module, via the API mould
Block using 120 and services the accessible various data and service for being provided or being safeguarded by social networking service.For example, using
API, application 120 can request and/or receive one or more navigation and recommend.Such application 120 can be based on browser
Application 120, or can be specific to operating system.Particularly, some applications 120 can be with Mobile operating system one
It rises and is resident on one or more mobile devices (for example, phone or tablet computing device) and executes (at least partly).This
Outside, although in many cases, can be the entity exploitation by operation social networking service using the application 120 or service of API
With maintenance using 120 and service, but any content in addition to data privacy concern will not all prevent API from being provided to public affairs
Certain third parties under many or special arrangement, so that navigational suggestion be made to can be used for third-party application 128 and service.
Although here search engine 216 is known as using in the context of social networking service, it is contemplated that it
It can also be used in the context of any website or online service.In addition, although the feature of the disclosure is referred to herein as
It uses or presents in the context of webpage, it is contemplated that any user interface views (for example, in mobile device or table
User interface on thin-skinned part) it is within the scope of this disclosure.
In the exemplary embodiment, it when members profiles are indexed, creates and stores sweep forward index.Search engine 216
Convenient for indexing and searching for the data or information that include in the content in social networking service, such as index and search data Layer, example
As profile data (being stored in such as profiles database 218), social graph data (are stored in such as social graph data library 220
In) and member activity and behavioral data (being stored in such as member activity and behavior database 222), and/or feeding pair
Information in image data library 224.In response to the received inquiry for information, search engine 216 can collect, parse
And/or the data in storage index or other similar structure, in order to identify and retrieve the information.This may include but unlimited
In sweep forward index, reverse indexing, N-gram index etc..
Feeding object database 224 may include can be in the feeding of one or more members of social networking service
The object of display.Feeding is the data format for providing a user the content of frequent updating.In social networking service, at
Member can check their feeding when they for example log in social networking service.Feeding is thought comprising social networking service can
It can make the interested one or more objects of user.User's feeding may include from different classes of entry, for example, position
Publication, user are issued, for suggestion, patronage model newly connected etc..Creation feeding means to from different classes of entry
It is ranked up, merges and feed displayed entries in order from different classes of entry, and creation user's feeding, the user.
The process being ranked up to entry and classification is usually extremely complex, since it is desirable that different targets, such as optimization are used
Family experience and from different classes of generation income.This process is usually very dull, and needs a large amount of experiment.
Although herein will feeding object database 224 be portrayed as comprising feeding object, it should be noted that need not by it is all this
A little potential feeding objects are aggregated in single database.In some example embodiments, as feeding object database 224
Substituted or supplemented, feeding object can be located in various other databases, and desire access to any component of feeding object
(such as search engine 216) can feed object across multiple database retrievals.However, for the sake of simplicity, this document will describe from
Feed the feeding object that object database 224 obtains.
Fig. 3 is the block diagram for illustrating in greater detail the application server module 214 of Fig. 2 according to example embodiment.Although
In many examples, application server module 214 will be various different dynamic comprising being used in social networking system 210 execute
The many sub-components made, but component those of related to the disclosure is depicted only in Fig. 3.
Here, see clearly platform (ingestion platform) 300 from profiles database 218, social graph data library 220,
Member activity and behavior database 222 and/or feeding object database 224 relevant to order models 302 obtain information.?
When training, sees clearly platform 300 and transmit this information to order models 302 to train order models 302, and in sequence, example
Such as when social networking service it needs to be determined that being seen clearly when which feeding object is presented to specific user and being presented with what sequence
Platform 300 will send information to order models 302, to allow the output of order models 302 to show in the feeding of user
It is various it is potential feeding objects sequences.
In some example embodiments, the information is sent in the form of feature vector.For example, each members profiles can be with
With what is formed by the information in profiles database 218, social graph data library 220 and member activity and behavior database 222
The feature vector of their own.In other example embodiments, platform 300 is seen clearly by raw information and is sent to order models 302,
And order models 302 create the feature vector of their own according to raw information.
Client interface server component 304 and the user interface client component 306 being located on client device 308 are logical
Letter is to run order models 302 and the feeding to user is shown or updated using its result.This can be inputted in response to user
It executes, user's input such as inputs the navigation for the webpage for including feeding.For example, user can indicate user interface
Client component 306 logs in social networking service account.Then the log-on message can be sent to client interface server group
Part 304, client interface server component 304 information can be used indicate to see clearly platform 300 from profiles database 218,
Letter appropriate is retrieved in social graph data library 220, member activity and behavior database 222 and/or feeding object database 224
Breath.
It is then possible to send client interface server component 304, user interface for the result from order models 302
Server component 304 can choose and format together feeding object appropriate with user interface client component 306 to show
To user.Be described in more detail below on how to via user interface client component 306 on client device 308
Show the details of these objects.
Fig. 4 is the block diagram for illustrating in greater detail the order models 302 of Fig. 3 according to example embodiment.In training assembly
In 400, sample members profiles 402, sample feeding object 403 and/or sample member activity and behavioural information 404 are input into
Feature extractor 406, feature extractor 406 be used for from sample members profiles 402, sample feeding object 403 and/or sample at
Member's activity and behavioural information 404 extract planning feature 408.It is characterized in the related variable of input with data.Due to showing some
Input can be data relevant to the member of social networking service in example embodiment, such as members profiles, member use and/
Or activity data or social graph information, therefore this feature can be such as member use and/or activity data or socialgram,
A part of member's personal information.This feature is also possible to the variable calculated according to a part of data, such as average, summation,
Difference, measured value etc..This feature can also be sample feeding object in some terms, such as title, the term frequently occurred,
And/or the various measurements about object, for example, the frequency of the appearance of keyword.
In the exemplary embodiment, planning feature 408 is then used as the input to machine learning algorithm 410, with training machine
Learning model 412 should show the probability of feeding object to generate to specific user.Although the probability can be with based on user
In some way with feeding object interaction a possibility that, as will be described in more detail, but it can also based on user or
The influence of the virus movement of downstream user, and these influence the relative worth to entire social networking service as a whole.?
In certain form of machine learning algorithm, training may include that sample results label 414 is supplied to machine learning algorithm 410.
Each of these sample results labels 414 are the possibility that instruction should show corresponding sample feeding object to user
The score of property.
Machine learning algorithm 410 can be selected from many different potential supervision or unsupervised machine learning algorithm.Prison
Superintend and direct machine learning algorithm example include artificial neural network, Bayesian network, instance-based learning, support vector machines, with
Machine forest, linear classifier, quadratic classifier, k- arest neighbors, decision tree and hidden Markov model.Unsupervised machine learning
The example of algorithm includes expectation-maximization algorithm, vector quantization and information bottleneck method.In the exemplary embodiment, using binary
Logic Regression Models.Dualistic logistic regression processing for dependent variable observation result can only there are two types of may type the case where.It patrols
It volume returns for predicting that a kind of situation or another situation are genuine possibility based on the value of independent variable (predictive variable).
In feeding object order engine 416, candidate's feeding object 418 is input into feature extractor 420, and feature mentions
Device 420 is taken to extract planning feature 422 from candidate's feeding object 418.Then, planning feature 422 is used as to machine learning model
412 input, what the output instruction of machine learning model 412 should show corresponding candidate feeding object 418 in feeding can
The score of energy property.
In the exemplary embodiment, training machine learning model 412 in such a way: it can export each potential feedback
Send the score of entry.The score generated by machine learning model 412 can be referred to as " it is expected that participation ".
Although the foregoing describe machine learning models 412 how to generate expected participation, in the exemplary embodiment, machine
Device learning model 412 is actually the combination of multiple models, especially world model's (also referred to as fixed-effect model) and one
Or the combination of multiple partial models (also referred to as random-effect model).In this way, training described in Fig. 4 can actually include
It include the difference training of the different models of machine learning model 412.
Fig. 5 depicts the machine learning model 412 of Fig. 4 according to example embodiment in more detail.Here, machine learning
Model 412 includes world model 500, every viewer's model 502 and every actor's model 504.Although not describing here,
In some example embodiments, additional each entry model can also be added, each potential feeding entry is corresponded to, or
The classification (for example, the publication of article, position, state update etc.) of at least each potential feeding entry.
The training data including the data about many different viewers and many different actors can be used to train
World model 500.The data of the different actors of the entry interacted about viewer can be used to train every viewer
Model 502.The data of the different viewers for the entry taken action to it about actor can be used to train every row
Dynamic person's model 504.Score combiner 506 can be used for for the score that model 500,502 and 504 exports being combined into single potential feedback
Send the single expected participation score of entry.It should be noted that score combiner 506 can assign a weighting to different model 500-
Each of 504, and these weights can be learnt via machine learning algorithm.Therefore, (example in some cases
Such as, certain features based on potential feeding entry, such as classification, position), it could possibly be higher than for the weight of every observer's model
For the weight of every actor's model, and in other cases, it is likely lower than for the weight of every viewer's model for every row
The weight of dynamic person's model.
Under certain technical situations, it is understood that there may be cold start-up problem, for example, due to lacking enough training datas, at random
There may be queries for the reliability of one of effect model 502,504.In order to solve the technical problem, threshold value can be used, wherein
The weight of specific random-effect model 502,504 can be set to 0, certain threshold quantity until obtaining training data.
At runtime, when it needs to be determined that each candidate feeds entry 418 when for one group of potential sequence for feeding entry
It can have the feature extracted from it and be input to machine learning model 412.The information may include about related potential feeding
The actor of entry and potential viewer information (although in some cases, about the actor in relation to potential feeding entry
Machine learning model 412 can be submitted to independently of candidate's feeding object 418 and/or each other with the information of potential observer).
Then each of the independent model 500-504 in machine learning model 412 can be entered relevant information into,
Each model exports the score that can be weighted by score combiner 506, then score combiner 506 by weighted score combine with
Output participates in score for specific potential the expected of feeding entry.Then the expected score that participates in can be used to come to potential feeding
Entry is ranked up and is determined and shown which potential feeding entry and be shown with what sequence.
The example of global characteristics includes being acted about the nearest click or virus for feeding entry (for example, sharing, liking
Deng), the network size of actor and/or viewer, actor/viewer's interaction, the impression based on different sliding windows, see
The person of seeing/actor support, viewer/actor " news value " score (, news value score maximally related with article
Generally indicate that this article has a news value more), feed position, the age for feeding entry, the network size of viewer, for seeing
Position, the viewer of job hunter's score a possibility that (instruction viewer's job search) of the person of seeing, viewer and/or actor
And/or the connection that actor has connect with time quantum, viewer and/or the actor since social networking service interaction is strong
Degree and the various profiles of viewer and/or actor, Activity Type, entry popularity, entry age etc..
It should be noted that it is only an example embodiment that the GAME type of multiple models, which uses,.In another example embodiment,
Using the A GLMix model with the fixed effect similar to world model, wherein stochastic effects are incorporated into for center
Every viewer and actor in GLMix model.In such a case, it is possible to by being based on each observerThe first subregion be then based on every actor Q={ Q1..., QACarry out subregion carry out parted pattern, wherein asking
Topic is represented as
Each training sampleFeature including three types, whereinFor learning global parameter,For
Learn the parameter specific to viewer, andFor learning the parameter specific to actor.It is also to be noted that this
The model of type can extend to other " each " classifications, such as each theme, each industry etc..
In the exemplary embodiment, it can train independently of one another each effect (global or local).For example, can instruct
Global and each actor model is repaired while practicing every viewer's model.In another example embodiment, global and each role
Model is off-line training, and viewer's model is trained in each viewing feeding every time.For each model, can incite somebody to action
The newly trained version of model is compared with older model, and old model is only replaced when detecting performance upgrade.
Fig. 6 is the screenshot capture according to user's feeding 600 including different classes of entry of some example embodiments.
In the example embodiment of Fig. 6, user's feeding 600 includes different classification, and 602,606 and of user's model are recommended in such as operation
Entry 608 is supported, and other embodiments may include additional categories.
In an example embodiment, user's feeding 600 provide and the job interest of user match and with use by oneself
602 (for example, job overalls 603 and 604) are recommended in the work that the particular job searching request at family is presented together.
User's model 606 includes the entry 607 (such as, the connection of user) issued by the user of social networking service, with
Comment is carried out to social networking service or including interested article or webpage.
Patronage entry 608 is the entry 609 placed by the sponsor of social networking service, is paid for presenting in user
The expense for issuing those entries is served, and may include that sponsor wants the advertisement for being generalized to webpage or link.
Although classification is shown as separation in user's feeding 600, can be mixed from different classes of entry
It closes, and is not only rendered as block.Therefore, user's feeding 600 may include a large amount of entries from each classification, and
The sequence of these entries is presented come certainly directional user for social networking service effectiveness based on expectations.
Fig. 7 is the stream for the method 700 using broad sense additivity Mixed effect model shown according to example embodiment
Cheng Tu.It is started the cycle in the graphic user interface of social networking service for each of multiple potential feeding entries
Feeding in show.For example, these potential feedbacks can be retrieved based on the searching algorithm of the viewer for graphic user interface
Send entry.At operation 702, first group of feature derived from the attribute of the first user in social networking service is obtained.It is grasping
Make at 704, obtain second group of feature derived from the attribute of the second user in social networking service, second user is to latent
In the actor of feeding entry.At operation 706, obtains about multiple feeding entries in social networking service from first and use
Third group feature derived from the activity at family.Operation 708 at, about in social networking service multiple feeding entries obtain from
4th group of feature derived from the activity of second user.
At operation 710, first group and second group of feature are input in the world model of machine learning, generate first
Expected the first computer based number participated in of user and potential feeding entry is estimated.At operation 712, by first group
It is input in every actor's model of machine learning with third group feature, generates the expection of the first user and potential feeding entry
The the second computer based number estimation participated in.At operation 714, second group and the 4th group of feature are input to engineering
In the every observer's model practised, the expected third computer based number participated in of the first user and potential feeding entry are generated
Word estimation.At operation 716, combination the first, second, and third computer based number estimation will to generate the first user
The estimation for a possibility that aprowl participating in the potential feeding entry in social networking service.
At operation 718, it is determined whether there are any more potential feeding entries.If it is, method 700 recycles
Operation 702 is returned to for next potential feeding entry.If it is not, then method 700 proceeds to operation 720.
At operation 720, based on the potential feeding entry that will aprowl be participated in social networking service to the first user
Activity a possibility that its corresponding estimation to be ranked up to potential feeding entry.At operation 722, based on sequence, scheming
The subset of multiple potential feeding entries is presented in shape user interface (GUI).
Fig. 8 is to show the block diagram 800 of software architecture 802, may be mounted at any one of above equipment or more
On a.Fig. 8 is only the non-limiting example of software architecture, and it should be recognized that many other frameworks may be implemented to promote
Into functions described herein.In various embodiments, software architecture 802 by the machine 900 of such as Fig. 9 etc hardware realization,
Including processor 910, memory 930 and input/output (I/O) component 950.In the exemplary architecture, software architecture 802 can
To be conceptualized as the storehouse of layer, wherein each layer can provide specific function.For example, software architecture 802 includes such as operating
The layer of system 804, library 806, frame 808 and application 810 etc.Consistent with some embodiments, operationally, application 810 passes through
Software stack calls API Calls 812 and calls 812 to receive message 814 in response to API.
In various implementations, operating system 804 manages hardware resource and provides public service.Operating system 804 includes example
Such as kernel 820, service 822 and driver 824.It is consistent with some embodiments, kernel 820 be used as hardware and other software layer it
Between level of abstraction.For example, kernel 820 provide memory management, processor management (for example, scheduling), assembly management, network and
Security setting and other function.Service 822 can provide other public services for other software layer.According to some implementations
Example, driver 824 are responsible for control or are engaged with bottom hardware.For example, driver 824 may include display driver, camera
Driver,OrLow energy driver, flash drive, serial communication are driven
Dynamic device (for example, universal serial bus (USB) driver),Driver, audio driver, power management driver
Deng.
In some embodiments, library 806 provides the rudimentary common base structure utilized by application 810.Library 806 may include
System library 830 (for example, C java standard library), can provide memory allocation function, character string operating function, mathematical function etc.
Function.In addition, library 806 may include API library 832, such as media library is (for example, for supporting the presentation of various media formats
With the library of manipulation, the various media formats such as mpeg-4 (MPEG4), advanced video coding (H.264 or
AVC), the 3rd layer of Motion Picture Experts Group (MP3), Advanced Audio Coding (AAC), adaptive multi-rate (AMR) audio coding decoding
Device, joint photographic experts group (JPEG or JPG) or portable network figure (PNG)), shape library is (for example, in display
On graphics environment in the OpenGL frame that is presented with two-dimentional (2D) and three-dimensional (3D)), data Kuku is (for example, for providing respectively
The SQLite of kind of relation data library facility), the library web (for example, for providing the WebKit of web browsing function) etc..Library 806 is also
It may include various other libraries 834, to provide many other API to application 810.
According to some embodiments, frame 808 provides the advanced common base structure that can be used by application 810.For example, frame
Frame 808 provides various GUI functions, advanced resource management, high-level position service etc..Frame 808 can provide can be by applying
Other API of 810 wide spectrums used, some of API can be specific to specific operation system 804 or platforms.
It in the exemplary embodiment, include domestic applications 850, contact application 852, browser application 854, book using 810
Nationality reader application 856, location application 858, media application 860, message transmission using 862, game application 864 and it is various its
He applies, such as third-party application 866.According to some embodiments, application 810 is to execute the program of function defined in program.
It can be created using various programming languages using one or more of 810, be constructed in various ways, the mode is all
If the programming language (for example, Objective-C, Java or C++) or procedural of object-oriented are (for example, C or compilation
Language).In particular example, third-party application 866 by the entity other than the supplier of particular platform (for example, used
ANDROIDTMOr IOSTMThe application of Software Development Kit (SDK) exploitation) it can be the shifting run on Mobile operating system
Dynamic software, the Mobile operating system such as IOSTM、ANDROIDTM、Phone or other moving operation systems
System.In this example, third-party application 866 can call the API Calls provided by operating system 804 812 to promote to retouch here
The function of stating.
Fig. 9 shows the graphical representation of the machine 900 of computer system form according to example embodiment, wherein can be with
One group of instruction is executed so that machine executes any one or more of process discussed herein.Specifically, Fig. 9 is shown
The graphical representation of machine 900 in the exemplary forms of computer system, wherein 916 can be executed instruction (for example, software, journey
Sequence, using 810, small application, app or other executable codes) for causing machine 900 to execute in process discussed herein
It is any one or more of.For example, instruction 916 can make the method 700 of the execution of machine 900 Fig. 7.Additionally or alternatively, refer to
Enable 916 Fig. 1-7, etc. may be implemented.General, unprogrammed machine 900 is transformed into specific machine 900 by instruction 916,
The machine 900 is programmed to execute described and illustrated function in the manner described.In an alternative embodiment, machine 900
It is operated as autonomous device or (for example, networking) can be coupled to other machines.In networked deployment, machine 900 can be
Run in server-client network environment with server machine or the ability of client machine, or as equity (or point
Cloth) peer machines operation in network environment.Machine 900 can include but is not limited to server computer, client calculates
Machine, PC, tablet computer, laptop computer, net book, set-top box (STB), portable digital-assistant (PDA), amusement matchmaker
System system, cellular phone, smart phone, mobile device, wearable device (for example, smartwatch), smart home device (example
Such as, smart machine), other smart machines, the network equipment, network router, the network switch, bridge or can sequentially or
Otherwise execute instruction 916 any machine, 916 movement to be taken of specified machine 900 of described instruction.Although in addition,
Individual machine 900 is illustrated only, but term " machine " should also include the set of machine 900, either individually or collectively execute
Instruction 916 is to execute any one or more of process discussed herein.
Machine 900 may include processor 910, memory 930 and I/O component 950, can be configured as example through
It is communicated with one another by bus 902.In the exemplary embodiment, processor 910 is (for example, central processing unit (CPU), reduced instruction set computer
Calculate (RISC) processor, complex instruction set calculation (CISC) processor, graphics processing unit (GPU), digital signal processor
(DSP), specific integrated circuit (ASIC), RF IC (RFIC), another processor or its any suitable combination) it can
Processor 912 and processor 914 including such as executable instruction 916.Term " processor " is intended to include multi-core processor,
The multi-core processor may include may be performed simultaneously instruction 916 two or more independent processors (sometimes referred to as
" core ").Although Fig. 9 shows multiple processors 910, machine 900 may include the single processor with single core,
Single processor (for example, multi-core processor) with multiple cores, multiple processors with single core, with multiple cores
Multiple processors or any a combination thereof.
Memory 930 may include main memory 932, static memory 934 and storage unit 936, all these all
For example may have access to via bus 902 to processor 910.Main memory 932, static memory 934 and memory cell 936
Storage embodies the instruction 916 of any one or more of method described herein or function.Instruction 916 can also its by
Machine 900 is completely or partially resided in during executing in main memory 932, static memory 934 is interior, storage unit 936
It is interior, at least one of processor 910 (for example, in the cache memory of processor) or any suitable combination
It is interior.
I/O component 950 may include various components to receive input, provide output, generation output, transmission letter
Breath, exchange information, capture measurement etc..Machine 900 will be depended on including the specific I/O component 950 in specific machine 900
Type.For example, the portable machine of such as mobile phone may include touch input device or other such input mechanisms,
And headless server machine may not include such touch input device.It should be recognized that I/O component 950 may include
Unshowned many other components in Fig. 9.I/O component 950 is grouped according to function and is only used for simplifying following discussion, and
And grouping is restrictive by no means.In various example embodiments, I/O component 950 may include output precision 952 and input
Component 954.Output precision 952 may include visible component (for example, the display of such as Plasmia indicating panel (PDP), hair
Optical diode (LED) display, liquid crystal display (LCD), projector or cathode-ray tube (CRT)), acoustic assembly (such as raises
Sound device), Haptics components (for example, vibrating motor, resistance mechanisms), alternative signal generator, etc..Input module 954 can be with
Including alphanumeric input module (for example, keyboard, be configured to receive alphanumeric input touch screen, optical keyboard or other
Alphanumeric input module), the input module based on point is (for example, mouse, touch tablet, trace ball, control stick, motion sensor
Or another direction instrument), tactile input module is (for example, physical button, providing and touching or the position of touch gestures and/or power
Touch screen or other tactile input modules), audio input component (for example, microphone) etc..
In further exemplary embodiments, I/O component 950 may include biometric component 956, moving parts
958, in environment components 960 or location component 962 and extensive other assemblies.For example, biometric component 956 can be with
Including component, be used to detect expression (for example, hand expression, facial expression, acoustic expression, body gesture or eyes tracking),
Bio signal (for example, blood pressure, heart rate, body temperature, perspiration or E.E.G), identification people are measured (for example, speech recognition, retina are known
Not, face recognition, fingerprint recognition or the identification based on electroencephalogram) etc..Moving component 958 may include acceleration transducer portion
Part (for example, accelerometer), gravity sensor component, rotation sensor component (for example, gyroscope) etc..Environment components 960 can
To include such as illumination sensor component (for example, photometer), temperature sensor assembly (for example, one of detection environment temperature
Or multiple thermometers), humidity sensor assemblies, pressure sensor assembly (for example, barometer), acoustics sensor device assembly (example
Such as, one or more microphones of ambient noise are detected), proximity sensor component (for example, detect nearby object infrared biography
Sensor), gas sensor is (for example, gas detection sensor, for detecting the concentration of hazardous gas to ensure safety or measurement
Pollutant in atmosphere) or can provide it is corresponding with surrounding physical environment instruction, measurement or signal other assemblies.Position
Component 962 may include position sensor assembly (for example, global positioning system (GPS) receiver assembly), height sensor group
Part (for example, altimeter or barometer that detection can be derived from the air pressure of height), direction sensing device assembly are (for example, magnetic
Power meter) etc..
Various technologies can be used to realize communication.I/O component 950 may include communication component 964, can operate with
Machine 900 is coupled to network 980 or equipment 970 via coupling 982 and coupling 972 respectively.For example, communication component 964 can be with
Including network interface components or another suitable equipment engaged with network 980.In other examples, communication component 964 can wrap
Include wire communication component, wireless communication components, cellular communication component, near-field communication (NFC) component,Component
(for example,Low energy),Component and other communication components are used to provide via other modes
Communication.Equipment 970 can be any one of another machine or various peripheral equipments (for example, the periphery coupled via USB
Equipment).
In addition, communication component 964 can detecte identifier or detect the component of identifier including that can operate.For example, logical
Believe that component 964 may include radio frequency identification (RFID) tag reader component, NFC intelligent label detection components, optical pickup
Component is (for example, the optical sensor for detecting one-dimensional bar code, the one-dimensional bar code such as Universial Product Code (UPC)
Codabar code, multi-dimensional bar code, as quick response (QR) code, Aztec code, Data Matrix, Dataglyph,
MaxiCode, PDF417, Ultra Code, UCC RSS-2D bar code and other optical codes) or Acoustic detection component
(for example, microphone of the audio signal marked for identification).Furthermore it is possible to various information are exported via communication component 964,
Such as via internet the position of agreement (IP) geo-location, via the position of Wi-Fi signal triangulation, via detecting
It can indicate the position, etc. of the NFC beacon signal of specific position.
Executable instruction and machine-storage medium
Various memories (that is, 930,932,934 and/or processor 910 memory) and/or storage unit 936 can be with
One or more groups of instructions 916 and data structure (for example, software) are stored, times in method described herein or function is embodied
What is one or more or by its utilization.When being executed by processor 910, these instructions (for example, instruction 916) make various behaviour
Make realization the disclosed embodiments.
As it is used herein, term " machine-storage medium ", " equipment storage medium " and " computer storage medium " table
Show identical things and may be used interchangeably.These terms refer to storage executable instruction 916 and/or data it is single
Or multiple storage equipment and/or medium are (for example, centralized or distributed database and/or associated cache kimonos
Business device).Therefore, term should be considered as including but not limited to solid-state memory and optics and magnetic medium, including processor
Memory inside or outside 910.Machine-storage medium, computer storage medium and/or the specific of equipment storage medium are shown
Example includes nonvolatile memory, including such as semiconductor memory devices, such as Erasable Programmable Read Only Memory EPROM
(EPROM), electrically erasable programmable read-only memory (EEPROM), field programmable gate array (FPGA) and flash memory device；Magnetic
Disk, such as internal hard drive and moveable magnetic disc；Magneto-optic disk；With CD-ROM and DVD-ROM disk.Term " machine-storage medium ",
" computer storage medium " and " equipment storage medium " clearly eliminates carrier wave, modulated data signal and other such Jie
Matter, wherein at least some are covered by term " signal media " described below.
Transmission medium
In various example embodiments, one or more parts of network 980 can be ad hoc network, Intranet, outer
Networking, VPN, LAN, WLAN, WAN, WWAN, MAN, internet, a part of internet, a part of PSTN, Plan Old electricity
Words service (POTS) network, cellular phone network, wireless network, Wi-Fi network, another type of network or two kinds or more
The combination of a variety of such networks.For example, a part of network 980 or network 980 may include wireless or cellular network, and
And coupling 982 can be CDMA (CDMA) connection, global system for mobile communications (GSM) connection or other kinds of honeycomb
Or wireless coupling.In this example, any one of various types of data transmission technologies may be implemented in coupling 982, such as
Single carrier radio transmission technique (1xRTT), Evolution-Data Optimized (EVDO) technology, general packet radio service (GPRS) skill
Art, the data transfer rate of enhancing for GSM evolution (EDGE) technology, third generation partner program (3GPP) comprising 3G, the
Four generations are wireless (4G) network, Universal Mobile Telecommunications System (UMTS), high-speed packet access (HSPA), the mutual behaviour of Worldwide Interoperability for Microwave access
The property made (WiMAX), long term evolution (LTE) standard are transmitted by various standards setting organizations, other remote protocols or other data
The other standards of technical definition.
Instruction 916 can be via network interface device (e.g., including the network interface components in communication component 964)
Pass through network using transmission medium, and using any one of multiple well-known transport protocols (for example, HTTP)
980 send or receive.Similarly, instruction 916 can be used transmission medium sent via coupling 972 (for example, equity couple) or
Receive equipment 970.Term " transmission medium " and " signal media " indicate identical things and can be interchanged in the disclosure
It uses.Term " transmission medium " and " signal media " should be considered as including that can store, encode or carry instruction 916 for machine
Device 900 execute any intangible medium, and including for promoting the communication of such software number or analog communication signal or
Other intangible mediums.Therefore, term " transmission medium " and " signal media " should be considered as including any type of modulation data letter
Number, carrier wave etc..It is set or changed in a manner of encoding to the information in signal in term " modulated message signal " expression
The signal of one or more of characteristic.
Computer-readable medium
Term " machine readable media ", " computer-readable medium " and " device-readable medium " indicates identical things, and
And it can be used interchangeably in the disclosure.These terms are defined to include both machine-storage medium and transmission medium.Cause
This, these terms include storage device/medium and carrier wave/modulated data signal.
Claims (20)
1. a kind of system for feeding to be presented in the graphic user interface for the calculating device display checked by the first user,
Include:
A kind of computer-readable medium, is stored thereon with instruction, described instruction when being executed by a processor so that the system is used
In:
For each of multiple potential feeding entries:
Obtain first group of feature derived from the attribute of first user in social networking service；
Obtaining second group of feature, the second user derived from the attribute of the second user in the social networking service is pair
The actor of the potential feeding entry；
About multiple feeding entries in the social networking service, the third group derived from the activity of first user is obtained
Feature；
About multiple feeding entries in the social networking service, obtain the 4th group derived from the activity of the second user
Feature；
First group of feature and second group of feature are input in the world model of machine learning, described first is generated and uses
Expected the first computer based number participated in of family and the potential feeding entry is estimated；
First group of feature and the third group feature are input in every actor's model of machine learning, generate described the
Expected the second computer based number participated in of one user and the potential feeding entry is estimated；
The second feature and the 4th group of feature are input in every observer's model of machine learning, generate described first
The expected third computer based number participated in of user and the potential feeding entry is estimated；And
In conjunction with the first computer based number estimation, the second computer based number estimation and the third base
In the number estimation of computer, with generate first user will aprowl participate in it is described latent in the social networking service
Estimation a possibility that feeding entry.
2. the system as claimed in claim 1, wherein every actor's model of the machine learning and the machine learning it is every
Observer's model is the subregion of the world model of the machine learning.
3. the system as claimed in claim 1, wherein the world model of the machine learning is fixed-effect model.
4. the system as claimed in claim 1, wherein every of every user model of the machine learning and the machine learning
Mesh model is random-effect model.
5. the system as claimed in claim 1, wherein described instruction also makes the system be based on first user will be in work
The estimation for a possibility that first entry in the social networking service is participated in dynamic puts the first entry to determine
It sets in the feeding of first user.
6. the system as claimed in claim 1, wherein described instruction also makes the system be based on first user will be in work
The estimation for a possibility that first entry in the social networking service is participated in dynamic is by the first entry at other
It is ranked up in possible entry to be serviced to first user.
7. the system as claimed in claim 1, wherein the world model of the machine learning, machine learning every actor's mould
Type and every observer's model of machine learning are respectively trained independently of one another.
8. a kind of for the calculating of feeding to be presented in the graphic user interface for the calculating device display checked by the first user
Machine method, comprising:
Obtain first group of feature derived from the attribute of first user in social networking service；
Obtaining second group of feature, the second user derived from the attribute of the second user in the social networking service is pair
The actor of potential feeding entry；
About multiple feeding entries in the social networking service, the third group derived from the activity of first user is obtained
Feature；
About multiple feeding entries in the social networking service, obtain the 4th group derived from the activity of the second user
Feature；
First group of feature and second group of feature are input in the world model of machine learning, described first is generated and uses
Expected the first computer based number participated in of family and the potential feeding entry is estimated；
First group of feature and the third group feature are input in every actor's model of machine learning, generate described the
Expected the second computer based number participated in of one user and the potential feeding entry is estimated；
Second group of feature and the 4th group of feature are input in every observer's model of machine learning, generate described the
The expected third computer based number participated in of one user and the potential feeding entry is estimated；And
Combine the first computer based number estimation, the second computer based number estimation and the third base
In the number estimation of computer, with generate first user will aprowl participate in it is described latent in the social networking service
Estimation a possibility that feeding entry.
9. according to the method described in claim 8, wherein, every actor's model of the machine learning and the machine learning
Every observer's model is the subregion of the world model of the machine learning.
10. method according to claim 8, wherein the world model of the machine learning is fixed-effect model.
11. according to the method described in claim 8, wherein, every user model of the machine learning and the machine learning
Every entry model is random-effect model.
12. method according to claim 8 further includes that will aprowl participate in the social networks based on first user
The estimation of the possibility of the first entry in service determines that the first entry, which is placed on described first, to be used
In the feeding at family.
13. method according to claim 8 further includes that will aprowl participate in the social networks based on first user
The first entry is dived at other and is carried out in the entry by the estimation of the possibility of the first entry in service
Sequence services the first entry based on the sequence with servicing to first user.
14. according to the method described in claim 8, wherein, the world model of the machine learning, machine learning every actor
Model and every observer's model of machine learning are respectively trained independently of one another.
15. a kind of non-transitory machinable medium including instruction, described instruction is worked as to be realized by one or more machines
When, so that one or more of machines are executed for graphical user circle in the calculating device display checked by the first user
The operation of feeding is presented in face, the operation includes:
Obtain first group of feature derived from the attribute of first user in social networking service；
Obtaining second group of feature, the second user derived from the attribute of the second user in the social networking service is pair
The actor of potential feeding entry；
About multiple feeding entries in the social networking service, the third group derived from the activity of first user is obtained
Feature；
About multiple feeding entries in the social networking service, obtain the 4th group derived from the activity of the second user
Feature；
First group of feature and second group of feature are input in the world model of machine learning, described first is generated and uses
Expected the first computer based number participated in of family and the potential feeding entry is estimated；
First group of feature and the third group feature are input in every actor's model of machine learning, generate described the
Expected the second computer based number participated in of one user and the potential feeding entry is estimated；
Second group of feature and the 4th group of feature are input in every observer's model of machine learning, generate described the
The expected third computer based number participated in of one user and the potential feeding entry is estimated；And
Combine the first computer based number estimation, the second computer based number estimation and the third base
In the number estimation of computer, with generate first user will aprowl participate in it is described latent in the social networking service
Estimation a possibility that feeding entry.
16. non-transitory machinable medium as claimed in claim 15, wherein every actor of the machine learning
Model and every observer's model of the machine learning are the subregions of the world model of the machine learning.
17. non-transitory machinable medium as claimed in claim 15, wherein the world model of the machine learning
It is fixed-effect model.
18. non-transitory machinable medium as claimed in claim 15, wherein every user's mould of the machine learning
Type and every entry model of the machine learning are random-effect models.
19. non-transitory machinable medium as claimed in claim 15, wherein the operation further includes based on described
First user by the estimation of the possibility of the first entry aprowl participated in the social networking service,
The first entry is placed in the feeding of first user by determination.
20. non-transitory machinable medium as claimed in claim 15, wherein the operation further includes based on described
First user by the estimation of the possibility of the first entry aprowl participated in the social networking service,
The first entry is dived at other and is ranked up in the entry to be serviced to first user, and is based on the row
Sequence services the first entry.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US15/900,219 | 2018-02-20 | ||
| US15/900,219 US10949480B2 (en) | 2018-02-20 | 2018-02-20 | Personalized per-member model in feed |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110175297A true CN110175297A (en) | 2019-08-27 |
Family
ID=67617940
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910121550.1A Pending CN110175297A (en) | 2018-02-20 | 2019-02-19 | Personalized every member's model in feeding |
Country Status (2)
| Country | Link |
|---|---|
| US (1) | US10949480B2 (en) |
| CN (1) | CN110175297A (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11030253B2 (en) * | 2018-03-02 | 2021-06-08 | Sap Se | Managing data feeds from different applications for users |
| CN109543066B (en) * | 2018-10-31 | 2021-04-23 | 北京达佳互联信息技术有限公司 | Video recommendation method and device and computer-readable storage medium |
| WO2021048627A2 (en) * | 2019-09-11 | 2021-03-18 | Adeption Limited | Systems and methods for context-based content generation |
| US11880376B2 (en) * | 2021-08-03 | 2024-01-23 | Hulu, LLC | Reweighting network for subsidiary features in a prediction network |
| US20230334514A1 (en) * | 2022-04-18 | 2023-10-19 | Microsoft Technology Licensing, Llc | Estimating and promoting future user engagement of applications |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120023043A1 (en) * | 2010-07-21 | 2012-01-26 | Ozgur Cetin | Estimating Probabilities of Events in Sponsored Search Using Adaptive Models |
| US20160358229A1 (en) * | 2015-06-05 | 2016-12-08 | Facebook, Inc. | Selecting Content for Presentation to Social Networking System Users Based On User Engagement with Content |
| US20170323268A1 (en) * | 2016-05-03 | 2017-11-09 | Linkedin Corporation | Scalable response prediction using personalized recommendation models |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9747647B2 (en) * | 2014-07-21 | 2017-08-29 | Facebook, Inc. | Missed feed unit for content feeds |
| US10083454B2 (en) * | 2015-01-30 | 2018-09-25 | Microsoft Technology Licensing, Llc | Social network content item federation based on item utility value |
| US20180089318A1 (en) * | 2016-09-23 | 2018-03-29 | Linkedin Corporation | Two-sided network growth optimization in social networks |
-
2018
- 2018-02-20 US US15/900,219 patent/US10949480B2/en active Active
-
2019
- 2019-02-19 CN CN201910121550.1A patent/CN110175297A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120023043A1 (en) * | 2010-07-21 | 2012-01-26 | Ozgur Cetin | Estimating Probabilities of Events in Sponsored Search Using Adaptive Models |
| US20160358229A1 (en) * | 2015-06-05 | 2016-12-08 | Facebook, Inc. | Selecting Content for Presentation to Social Networking System Users Based On User Engagement with Content |
| US20170323268A1 (en) * | 2016-05-03 | 2017-11-09 | Linkedin Corporation | Scalable response prediction using personalized recommendation models |
Also Published As
| Publication number | Publication date |
|---|---|
| US10949480B2 (en) | 2021-03-16 |
| US20190258741A1 (en) | 2019-08-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10990899B2 (en) | Deep and wide machine learned model for job recommendation | |
| US10860670B2 (en) | Factored model for search results and communications based on search results | |
| US11372940B2 (en) | Embedding user categories using graphs for enhancing searches based on similarities | |
| US20180285824A1 (en) | Search based on interactions of social connections with companies offering jobs | |
| US11163845B2 (en) | Position debiasing using inverse propensity weight in machine-learned model | |
| US11204973B2 (en) | Two-stage training with non-randomized and randomized data | |
| US10607189B2 (en) | Ranking job offerings based on growth potential within a company | |
| CN110175297A (en) | Personalized every member's model in feeding | |
| CN109978175A (en) | Parallelization coordinate descent for machine learning model | |
| US11514115B2 (en) | Feed optimization | |
| US11151661B2 (en) | Feed actor optimization | |
| US20220284068A1 (en) | User electronic message system | |
| US20180218328A1 (en) | Job offerings based on company-employee relationships | |
| US10572835B2 (en) | Machine-learning algorithm for talent peer determinations | |
| US20180285822A1 (en) | Ranking job offerings based on connection mesh strength | |
| US11397742B2 (en) | Rescaling layer in neural network | |
| CN110427564A (en) | Deep learning is integrated into broad sense additivity melange effect (GAME) frame | |
| US11334612B2 (en) | Multilevel representation learning for computer content quality | |
| CN110955840B (en) | Joint optimization of notifications and pushes | |
| US20240119278A1 (en) | Transfer learning for seniority modeling label shortage | |
| US20230075600A1 (en) | Neural network prediction using trajectory modeling | |
| US11816636B2 (en) | Mining training data for training dependency model | |
| US20190362013A1 (en) | Automated sourcing user interface | |
| US20230077840A1 (en) | Machine learning model for specialty knowledge base | |
| US11461421B2 (en) | Techniques for suggesting skills |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |